1

What to expect

The outcome in plain words before you begin

Here's what you're building

Right now, every time you open a new chat you're starting from zero. Your AI doesn't know your clients, your rates, your voice, or how you make decisions. You re-explain everything. It forgets.

After this setup, that's over. Your AI will carry your context forward — permanently. It knows who your biggest clients are, how you prefer to work, what you never compromise on, and what your business looks like today. You open a fresh chat and it already knows you.

Think of it like hiring an assistant who actually reads your files and remembers what you told them last week. Except they're available 24/7 and never forget.

The three things this sets up

  • Memory — a structured set of files that hold your key business facts, client notes, past decisions, and preferences. Your AI reads them automatically every session.
  • Rules & voice — a plain-text document (called CLAUDE.md) that tells your AI how you think, your values, what you never do, and how you want it to sound when it speaks for you. It's your operating manual.
  • Starter skills — 2–3 reusable workflows for the things you do most (draft a proposal, run a weekly check-in, brief yourself on a client before a call). You can always add more later.
Promise
You will not need to touch any code. Everything here is plain text and conversation. If you can write an email, you can do this.

2

Step-by-step setup

The simple way — about 5 minutes, no downloads
The easy way — no files to manage
You don't download or arrange anything. You connect an empty folder, paste one prompt, and your AI builds the whole kit itself. About 5 minutes, no code.
1

Where — connect an empty folder

Make a brand-new empty folder on your computer — a plain local one (ideally not inside OneDrive; e.g. C:\ai-memory). In Claude/Cowork, connect it and mark it Trusted so it stops asking permission every time.

2

What — paste this one prompt

Open a chat and paste the whole prompt below. Your AI builds its own rulebook, memory system, and skills — right in your folder.

Set up my "AI Memory" system in this connected folder. Create: 1) CLAUDE.md — my AI's rulebook: who I am / what my business does; how to act (know me from my memory, use my voice, just help on small reversible things, always be honest, never make things up); and a HARD LIMITS list of things you must never do without asking (send money, message a client, delete anything, post publicly). Add a map to memory/ and skills/, and this rule: whenever you learn a durable fact about me, save it to memory/ and add a line to memory/MEMORY.md. 2) memory/ — a MEMORY.md index (one line per fact) plus one example memory file. 3) skills/ — four SKILL.md files: remember, daily-brief, janitor (weekly tidy), wrap-up (end-of-session save). Then confirm what you built and interview me with four quick questions, one at a time: (1) what my business does + who it serves, (2) my top 1-3 clients, (3) my voice + hard limits, (4) two or three rules I repeat. Save each answer to memory/ and update MEMORY.md. Then tell me I am ready and to open a fresh chat to test it.
3

How — answer 4 quick questions

It interviews you and writes everything to memory itself. That's the whole setup — you never touch a file. Then it tells you to open a fresh chat and test it (next section).

Prefer the raw files? (optional)
A downloadable "raw files" version is included as a bonus. If you'd rather have the files directly: unzip it into a plain local folder (not OneDrive) and connect that folder — it already contains everything. You don't need to; the one prompt above is the easy path.

3

How to test it — the wow moment

Open a fresh chat and watch it just know

Once the build is done, close everything and open a completely new chat — no context, no history. Don't say anything about your business. Just ask one of the questions below.

If it answers correctly with context you only told it during the intake, the system is working.

Test prompt 1
"Who are my most important clients right now, and what do I need to remember about them?"
It should name them, what they care about, and anything you flagged as important — without you saying a word.
Test prompt 2
"Draft a quick proposal intro for a new client in [your industry]. Keep it in my voice."
It should write in your tone — not generic AI-speak. If it sounds like you, the voice calibration worked.
Test prompt 3
"What are my pricing rules? What do I never discount below?"
It should recall exactly what you said in the intake — your real floor, not a guess.
What if it gets something wrong?
That's normal and easy to fix. Just say: "That's not right — update this: [the correct info]." It will rewrite the memory file and remember it from that point forward. You're training it in real conversation, not code.

4

How to customize it

Mostly, you just talk to it

Conversation is the primary interface

You don't need to open any files to update your AI's knowledge. The simplest path is always: just tell it.

  • "Remember that my biggest client is Apex and they never want to be contacted on Fridays."
  • "Update my rates — my minimum project fee is now $3,000."
  • "Add a rule: I never agree to scope changes without a written amendment."

Each time you say something like this, it writes it to the right memory file automatically. Nothing to manage.

Optional: edit the files directly

If you want to go deeper, every part of the system is plain-text markdown. You can open the files in:

  • Obsidian — free app, beautiful for this kind of linked notes system
  • VS Code — if you're comfortable with a code editor
  • Any text editor — even Notepad works. The files are just text.

But you will never have to open a file. Conversation first, always.

Pro tip — the fastest customization
At the end of any session where you made an important decision, just say: "Save today's key decisions to memory." It will extract and store them for you.

5

How it grows with you

Small, consistent additions compound over time
🧠

Add memories as the business changes

New client? New rate? New rule you're living by? Just tell it. The memory system is append-and-update — it never gets confused by new information, it just stays current.

Add skills for new recurring tasks

Once you notice you're asking for the same thing week after week, that's a skill candidate. Say: "Turn what we just did into a reusable skill I can call next time." It builds it and saves it. Next time, one sentence triggers the whole workflow.

🔄

Let the weekly janitor do its job

The maintenance sweep runs automatically each week. It catches stale information, reorganizes what's grown messy, and keeps the system fast. You don't need to touch it — just notice if your morning brief or first response feels fresher each week. That's the janitor working.

📅

Suggested cadence: the 5-minute Friday check-in

Once a week (Friday works well), open a chat and say: "Quick check-in. What should I update in my memory before next week?" It will ask you three short questions and update itself. That's it. Five minutes keeps it razor-sharp.

Pro tip — the compounding effect
At week 1 it knows your basics. At month 1 it knows your patterns. At month 3 it's anticipating what you need before you ask. The system compounds — the work you put in now multiplies every week forward.

6

You're set.

What good looks like after 30 days

Your AI actually knows you now.

You've done the work once. From here, the system earns it back a little every single day.

No more re-briefing. No more starting from scratch. Just open a chat and go.

What good looks like at 30 days

  • You open a new chat and it greets you with the right context — without prompting.
  • First drafts sound like you, not like a generic AI.
  • You catch yourself saying "remember this" instead of re-explaining — and it just works.
  • You've added at least one new skill for something you kept repeating.
  • The weekly janitor has run 4 times. You haven't had to touch a file.
  • The time you used to spend re-briefing is time you're spending on actual work.